Code Llama on Apple M2 Ultra · 192GB

✅ Runs comfortably — Code Llama 34B @ Q8_0
Code Llama 34B at Q8_0 needs ~36.6 GB (weights 33.4 GB + KV 1.5 GB + overhead 1.7 GB @ 8K ctx) of your 141 GB usable (of 192 GB unified memory) — plenty of headroom. Expect ~7–11 tok/s (usable).
context @ Q8_0: comfortable to 16K
quantneedsspeed
✓ Q8_036.6 GB~7–11 tok/s (usable)
✓ Q5_K_M25.0 GB~10–16 tok/s (usable)
✓ Q4_K_M21.5 GB~11–19 tok/s (usable)
✓ Q3_K_M17.6 GB~14–25 tok/s (usable)
$runlocal install codellama:34b
runlocal verdict card
download card share on x share on reddit drop it in your model-drop post
Code Llama: Runs comfortably
for your own README — links back here

Same model, other machines: Apple M1 · Apple M1 Pro · Apple M2 · Apple M3 Pro · Apple M2 Pro · Apple M4 Pro

Also runs on a Apple M2 Ultra: DeepSeek-R1 (distill) · Qwen3 · Qwen2.5 · Llama 3.1 · Llama 3.3

Best models for a Apple M2 Ultra · 192GB →

Architecturally similar

These share Code Llama's KV-cache geometry — they size memory the same way, so here's how they fit on a Apple M2 Ultra · 192GB too:

modelsharesverdict here
Qwen2.5-Coder 8kv/128hd ✅ Runs comfortably
Llama 3.2 8kv/128hd ✅ Runs comfortably
Mistral 7B 8kv/128hd ✅ Runs comfortably
Mistral Small 3 8kv/128hd ✅ Runs comfortably
Mistral Nemo 8kv/128hd ✅ Runs comfortably
Codestral 8kv/128hd ✅ Runs comfortably
Llama 3.2 Vision 8kv/128hd ✅ Runs comfortably
Mixtral 8x7B 8kv/128hd ✅ Runs comfortably

…and 8 more — see the Code Llama page.

Check any combo yourself: open the checker →